Nemotron 3.5 Super VL
Neural Network
Nemotron 3.5 Super VL is an open multimodal NVIDIA model with MoE architecture for text, images, documents, and video.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 000 000 tokens
- Max answer length
- 131 072 tokens
- Reasoning effort
- low, medium, high
- Capabilities
- reasoning, function calling
How it works Nemotron 3.5 Super VL?
Nemotron 3.5 Super VL is an open multimodal NVIDIA model with a mixture-of-experts architecture: out of 120 billion parameters, 12 billion are active per request, combining large network capacity with efficient computation. It accepts text, images, documents, interface screenshots, and video, making it ideal for mixed data inputs like pages, frames, and screens. In BotHub, the model is accessible from Russia without a VPN or foreign card; payment is in rubles per token, and unused balances do not expire. Conversations are encrypted via AES-GCM, requests go through a unified OpenAI-compatible API, and you can switch models without changing integrations. For teams, we offer contracts, invoices, electronic document management (EDO), and an admin panel with limits. Use cases include: parsing contract and report scans to extract table data, testing application screens or agent scenarios, summarizing video recordings, answering questions about charts and diagrams, and comparing Nemotron's responses with other models in one window.Frequently asked questions about Nemotron 3.5 Super VL
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
The NVIDIA model works with text, documents, and images: it parses large files, answers content-related questions, helps with code (refactoring and debugging), and solves math and other STEM problems. A one-million token context allows you to keep an entire codebase or a large report in a single conversation.
The output limit is up to 131,072 tokens, which is the size of a small book. It is enough for detailed documentation, a large code module, or a comprehensive document analysis without splitting. If the response hits the limit, just ask the model to continue from where it stopped.
Yes, the model's reasoning before answering is confirmed: it thinks through the solution process first and then provides the result. This significantly helps with math, logic, multi-step data analysis, and debugging, where it is important to trace the entire chain of steps rather than just guessing the answer.
Yes, both function calling and structured JSON output based on a schema are supported. You can connect the model to your services, databases, and internal tools. In BotHub, everything goes through a unified OpenAI-compatible API, so you can switch models without rewriting your integration.
Images and documents — yes: send a screenshot, diagram, chart, or file, and the model will parse the content, describe details, and answer questions about it. We have no data on audio or video processing, so we do not promise such analysis.
Our data does not specify the model's language capabilities, so we cannot promise specific quality for Russian — it is more reliable to test it on your typical tasks. Access to Nemotron is open from Russia without a VPN or foreign card; payment is via Russian cards per token.